How we know what we know
Learning Curve maps the developmental terrain of childhood by cross-referencing 15 developmental frameworks from 66 countries, peer-reviewed research, and the constructs measured by leading assessment instruments. The result: a single canonical skill ontology, with every claim traceable to its sources.
- 2,135Canonical skills
- 1,011Milestones
- 15Frameworks
- 66Countries
- 2,032Voice activities
Reading paths
I want to understand my child
I use developmental frameworks daily
I want to see the rigor
Four layers of evidence
Every milestone and skill we track is informed by four overlapping layers of evidence. Where the layers agree, we have high confidence. Where they disagree, we surface the spread rather than picking a winner.
Evolutionary baselines
The deepest layer: developmental patterns that hold across human cultures and through evolutionary history. Sleep architecture, attachment windows, daily movement needs, and other patterns that show up wherever we look. These rarely change with parenting style.
Biology and developmental research
Peer-reviewed research on brain development, motor acquisition, and cognitive milestones. Each finding is cited to a specific paper, so you can follow the evidence back to the source.
Developmental frameworks and curricula
National guidelines and curricula from 66 countries: EYFS (England), Head Start ELOF (USA), Chinese 3–6 Guide, MHLW (Japan), and many more. When frameworks from different continents independently describe the same milestone at the same age, that’s likely biology rather than culture.
Parent-reported experience
Common patterns parents report — sleep regressions, picky eating, school-readiness anxiety — each paired with a research-grounded reframe. We cite the underlying findings rather than aggregate anonymous forum posts.
Here’s what those four layers produce: 2,135 skills you can browse, grouped by 11 developmental domains.
Why every skill has a confidence score
We score confidence two ways: how many independent sources agree (cross-framework agreement), and how cleanly each source extracted into structured data (extraction confidence). Both are honest about limits.
If twelve frameworks describe pretend play with stuffed animals between 18 and 30 months, that’s high agreement — likely a human-universal pattern. If only one curriculum lists a particular knowledge fact, that’s low agreement — probably culturally shaped or interest-driven.
High: agreement across most frameworks plus biology layer. Medium: documented across several frameworks, with some age variation. Low: emerging or framework-specific; treat with caution.
Whose childhood is normal?
Most developmental research has historically drawn on Western, educated, industrialised, rich, democratic samples — the WEIRD problem. We counter that by including frameworks from East Asia (China, Japan, South Korea, Hong Kong, Singapore), South Asia (India), Oceania (Australia, New Zealand), the Nordics, and varied Western traditions.
We do not claim cultural universality. We surface where frameworks converge — likely human-universal patterns — and where they diverge — culturally shaped practice. Parents should read age windows as ranges, not deadlines, and weight cultural context accordingly.
Children develop in clusters, not checklists
A child does not learn to roll, then to sit, then to crawl as isolated tasks completed one at a time. Those abilities emerge together — drawing on the same underlying postural control, visual tracking, and reach. Move one of them and the others shift. Developmental psychology calls these co-emerging sets developmental clusters, and they are the unit Learning Curve plans around, not individual checklist items.
This matters because activities that touch multiple capabilities at once — building a den with a sibling, helping cook a meal, telling a story with drawings — produce deeper and longer-lasting learning than narrow drills aimed at a single skill. The research is consistent across motor, cognitive, and social-emotional domains.
What the research says
- Developmental cascadesMasten & Cicchetti (2010) — early capabilities in one domain predict whole clusters of later capabilities across domains. The cluster, not the individual milestone, is the unit of cascading developmental effect. Development and Psychopathology 22(3): 491–495.
- Dynamic systems theoryThelen & Smith (1994) — motor abilities (rolling, sitting, crawling, cruising, walking) emerge from synchronous maturation of postural, perceptual, and motivational subsystems, not as isolated checklist achievements. A Dynamic Systems Approach to the Development of Cognition and Action (MIT Press).
- Cognitive ability clustersMcGrew (2009) — the Cattell-Horn-Carroll model shows cognitive abilities reliably cluster into broad and narrow factors across populations and ages. Intelligence 37(1): 1–10.
- Why standard assessments use clustersBayley-4, Vineland-3, and Mullen Scales all organise items by domain cluster — because the underlying construct each instrument measures is the cluster, not any single item within it.
- Cross-cluster practice outperforms drillRohrer & Pashler (2007) — interleaved practice across related capabilities produces better retention and transfer than blocked single-skill drill. Memory & Cognition 35(8): 1816–1822. Replicated in Brown, Roediger & McDaniel (2014) Make It Stick (Harvard University Press) and in Bransford et al (2000) How People Learn (National Academies Press).
- Representational redescriptionKarmiloff-Smith (1992) — developmental change is representation-redescription across clusters of related processes, not isolated module updates. Beyond Modularity (MIT Press).
How Learning Curve uses clusters
Every milestone in our system belongs to one or more developmental clusters. We favour activities that support multiple cluster members at once over narrow single-skill drills. Cluster boundaries are visible on every milestone's wiki page and on the mapping view in this methodology section. They are derived from a combination of: overlapping age windows within the same subcategory, shared prerequisites in the developmental graph, and cascade research that names specific co-emerging sets.
We do not claim our clusters are the only valid grouping. Where cascade research is contested, we present the alternative groupings side by side rather than picking a winner.
What our voice activities measure
We have 2,032 voice-coached activities, each designed to surface specific developmental signals during ordinary moments — storytime, mealtime, play. They are not clinical tests. They are structured observations a parent can do with their child, with the agent watching for evidence of specific skills.
What we capture, by domain
- Languagevocabulary breadth, sentence length, narrative coherence, comprehension
- Numeracycounting, ordinality, conservation, early operations, problem-solving talk
- Movementgross & fine motor, balance, bilateral coordination, body awareness
- Thinkingattention, working memory, cause-and-effect reasoning, planning
- Characterself-regulation, persistence, emotional resilience, value-formation
- Socialturn-taking, theory of mind, cooperation, conflict resolution
- Creativepretend play, drawing, music, story-making, dramatic arts
- Academicinquiry methods, evidence evaluation, source credibility, synthesis
- Natureobservation of living systems, environmental stewardship, life cycles
- Practicalself-care routines, household contribution, real-world skills
- Contemplativeattention training, reflective practice, gratitude, present-moment focus
How our skills map to assessment instruments
Our skill ontology is mapped to constructs measured by WHO GSED v2, Stanford SHQ, CREDI, Denver II, EYFS Profile, MELQO, Bayley-4, ASQ-3 (24mo), WHO MGRS, M-CHAT-R/F, UNICEF ECDI 2030, MICS6 ECDI. We map to the same developmental areas these instruments cover so what we track aligns with what professionals assess. We never reproduce their items, scoring rubrics, or administration protocols.
| Instrument | Type | Construct mappings |
|---|---|---|
| WHO GSED v2 | Normative developmental instrument | 310 |
| Stanford SHQ | Normative longitudinal study | 422 |
| CREDI | Normative developmental instrument | 108 |
| Denver II | Clinical screening | 78 |
| EYFS Profile | Curriculum profile | 51 |
| MELQO | Normative instrument | 30 |
| Bayley-4 | Clinical developmental assessment | 30 |
| ASQ-3 (24mo) | Developmental screening | 25 |
| WHO MGRS | Normative growth reference | 12 |
| M-CHAT-R/F | Autism screening | 11 |
| UNICEF ECDI 2030 | Normative instrument | 8 |
| MICS6 ECDI | Population survey (item mapping in progress) | 0 |
Normative calibration
Our milestone windows are derived from two complementary sources: expert-authored curriculum frameworks from 66 countries, and empirical child-response data from population surveys. The empirical layer includes responses from 371,670 children across 66 countries — drawn from the UNICEF MICS6 survey (2019–2023) and a US longitudinal study from Stanford (Stenhaug, Ram & Frank 2021). Where a milestone is supported by cross-cultural child-response data, its window reflects population-level acquisition age rather than solely expert opinion.
371,670
children with response data
MICS6 · Stanford SHQ · CREDI · WHO MGRS · IDB Bogota
66
countries represented
Dominated by UNICEF MICS6 (66 countries, 2019–2023)
15
research sources mapped
WHO GSED · Stanford SHQ · CREDI · Denver II · EYFS · +10 more
Browse the full dataset registry →·Read our product decisions →
What we're still working on
The methodology page is honest about what isn't finished.
- indexingPer-framework verbatim quotes and page-number citations on every skill, so you can read the original source language alongside our extraction.
- scopingAdditional assessment instruments beyond the four mapped today (Bayley-4, ASQ-3, EYFS Profile, M-CHAT-R/F).
- scopingExpanded coverage of underrepresented regions, especially African and Latin American developmental frameworks.
- in progressLongitudinal validation — comparing voice-activity observations against later professional assessments to refine our confidence scoring.
- in progressRefresh cadence for parent-experience data so reframes track current research rather than going stale.
What the data revealed
Building this map surfaced patterns we weren't looking for: why formal education crowds out character and movement, how domains emerged from the data rather than theory, why developmental science has a blind spot after age 5, and how learning relationships change from parent to peer to self across childhood.
Read the research findings →What this is not
Common questions
Is this a substitute for my pediatrician’s developmental check?
No. Learning Curve is informed by published developmental research, but it is not a clinical assessment and does not replace professional evaluation. Use it to understand your child’s journey and to have better-informed conversations with your pediatrician or health visitor. If you have specific concerns, see a clinician.
Why do you cite age ranges instead of fixed ages?
Children develop on different timelines. The age windows you see come from cross-referencing developmental guidance across 16 countries; where countries agree, the window narrows; where they diverge, the window widens. Treat ages as ranges, not deadlines.
How is this different from ASQ-3 or Bayley-4?
ASQ-3 is a developmental screen and Bayley-4 is a clinical assessment, both administered by professionals with validated scoring. Learning Curve is not a clinical instrument. Our skill ontology is mapped to constructs measured by these tools so that what we track aligns with what professionals assess, but we never reproduce their items, scoring rubrics, or administration protocols.
My child is ahead or behind on some skills — should I worry?
Variation is normal and expected. A child can be early on language and later on motor skills, or vice versa. The confidence band on each skill shows how much frameworks agree on the typical age range — a wide band means timing varies a lot. Persistent or widespread concerns deserve a conversation with a clinician.
How often do you update the framework data?
New frameworks and research findings are added as we ingest and validate them. Each update re-runs cross-framework agreement scoring and refreshes the skill ontology. The numbers you see on this page reflect the most recent verified state; they are dated next to the underlying constants in our codebase.
Where does the parent-difficulty data come from?
Layer 3 contains 81 documented patterns parents commonly report — things like sleep regressions, picky eating, or anxiety about reading readiness — each paired with a research reframe. The entries are sourced from peer-reviewed and clinical literature; we cite the underlying findings rather than aggregate anonymous forum posts.
Why are some domains better-covered than others?
Frameworks differ in what they emphasize: motor and language are well-mapped across most curricula, while areas like contemplative practice or cultural knowledge are covered by fewer sources. Coverage gaps are visible on the developmental Gantt and acknowledged in our ongoing-research section.